This is the official source code of the paper: "An Evasion Attack against Stacked Capsule Autoencoder".
- Author: Jiazhu Dai, Siwei Xiong
- Institution: Shanghai University
- Email: daijz@shu.edu.cn (J. Dai)
Code files under /capsules folder were copied from akosiorek/stacked_capsule_autoencoders before 2020/9/24.
Executable .py files:
- train.py: Train the SCAE model and save it under /checkpoints folder. Dataset cache files are saved under /datasets folder.
- test.py: Test the model under /checkpoints folder, generate and save the K-Means classifier at the same place.
- attack_gdu.py: Launch the attack with GDU algorithm using the model under /checkpoints folder. Results are saved under /results/gdu folder.
- attack_psc.py: Launch the attack with PSC algorithm using the model under /checkpoints folder. Results are saved under /results/psc folder.
- attack_opt.py: Launch the attack with OPT algorithm using the model under /checkpoints folder. Results are saved under /results/opt folder.
Please feel free to use it as you like.